Monocular Vision Based Distance Measurement of Frontal Obstacles for Unmanned Aerial Vehicle

The autonomous obstacle avoidance is one of the key techniques for UAV (Unmanned Aerial Vehicle) applications in large outdoor scenarios. However, the distance between the obstacle and UAV cannot be accurately estimated, since the 6-D pose of the UAV and the feature points on the obstacle boundary are difficult to obtain during the flight. In this paper, an obstacle distance estimation scheme based of the local geometric constraints is proposed. First, the local geometric constrains is introduced to the pipeline of pose estimation for UAV, which aims to improve the feature matching result by rejecting the outliers. Second, the proposed method used the local geometric constraints to find the robust correspondence pixels on the obstacle surface, which are essential factors for triangulation distance measurement. The experiment results show that the proposed scheme cannot only improve the 6-D pose estimation of the UAV, but can obtain a precise obstacle distance measurement, specifically the distance between obstacle boundary and the UAV.